OntoNotes A Unified Relational Semantic Representation.pdf

OntoNotes: A Unified Relational Semantic Representation Sameer S. Pradhan BBN Technologies Cambridge, MA 02138 Eduard Hovy ISI/USC Marina Del Rey, CA 90292 Mitch Marcus University of Pennsylvania Philadelphia, PA 19104 Martha Palmer University of Colorado Boulder, CO 80309 Lance Ramshaw BBN Technologies Cambridge, MA 02138 Ralph Weischedel BBN Technologies Cambridge, MA 02138 Abstract The OntoNotes project is creating a corpus of large- scale, accurate, and integrated annotation of multiple lev- els of the shallow semantic structure in text. Such rich, integrated annotation covering many levels will allow for richer, cross-level models enabling significantly better au- tomatic semantic analysis. At the same time, it demands a robust, efficient, scalable mechanism for storing and ac- cessing these complex inter-dependent annotations. We de- scribe a relational database representation that captures both the inter- and intra-layer dependencies and provide details of an object-oriented API for efficient, multi-tiered access to this data. 1 Introduction The OntoNotes project is addressing the challenge of large-scale, accurate, and integrated annotation of multiple levels of the shallow semantic structure in text. Experience has shown that when individual levels like syntactic parse structure, propositional structure, and semantic role labels can be annotated with high consistency (inter-tagger agree- ment, or ITA), then machine learning models can also be implemented to predict those structures with only somewhat lower accuracy. The premise of OntoNotes is that integrated annotation covering many levels will allow for richer, cross- level models enabling significantly better automatic seman- tic analysis. Given that goal, the mechanism used to store and access the annotation becomes a key part of the research, and not just a tool support issue. The representation has to correctly capture the dependencies between the different annotation layers, so that consistency can b